app.py

import functools as ft
import gradio as gr

import torch
import transformers
from transformers import T5ForConditionalGeneration, T5Tokenizer

tokenizer: T5Tokenizer = T5Tokenizer.from_pretrained(
    "models/roborovski/superprompt-v1", local_files_only=True
)
model: T5ForConditionalGeneration = T5ForConditionalGeneration.from_pretrained(
    "models/roborovski/superprompt-v1", local_files_only=True
)

@ft.lru_cache(maxsize=1024)
def super_prompt(text: str, seed: int, max_new_tokens: int, prompt: str) -> str:
    transformers.set_seed(seed)

    if max_new_tokens <= 0:
        max_new_tokens = 150

    with torch.inference_mode():
        if prompt:
            input_text = f"{prompt} {text}"
        else:
            input_text = f"Expand the following prompt to add more detail: {text}"

        input_ids = tokenizer(input_text, return_tensors="pt").input_ids

        outputs = model.generate(
            input_ids,
            max_length=max_new_tokens,
            do_sample=True,
            temperature=0.7,
            top_k=50,
            top_p=0.95,
        )

        return tokenizer.decode(outputs[0], skip_special_tokens=True)

demo = gr.Interface(
    fn=super_prompt,
    inputs=["text", "slider", "slider", "text"],
    outputs=["text"],
)

demo.launch()